Naval Surface Ship Design
Optimization for Affordability
PHASE
I
Grant: N00014-02-1-0796
Sponsor: Katherine Drew, ONR 334, Office of Naval
Research
Dr. Alan Brown and Dr. Wayne Neu, Virginia Tech
Naval ship concept design is
traditionally an “ad hoc” process.
Selection of design concepts for assessment is guided primarily by
experience, design lanes, rules-of-thumb and imagination. Communication and coordination between design
disciplines (hull form, structures, resistance, etc.) requires significant
designer involvement and effort. Concept
studies continue until resources or time run out. Critical elements missing from this process
are:
·
A consistent format and methodology for multi-objective decisions based
on dissimilar objective attributes: specifically effectiveness, cost and
risk. Mission effectiveness, cost and
risk cannot logically be combined as in commercial decisions where discounted
cost can usually serve as a suitable single objective. Multiple objectives must be presented
separately, but simultaneously, in a manageable format for trade-off and decision
making.
·
An efficient and robust method to search the design space for optimal
concepts.
·
Practical and quantitative methods for measuring effectiveness. An Overall Measure of Effectiveness (OMOE)
model or function is an essential prerequisite for optimization and design
trade-off. This effectiveness can be
limited to individual ship missions or extend to missions within a task group
or larger context.
·
Practical and quantitative methods for measuring risk. Overall risk includes schedule, production,
technology performance and cost factors. It is measured using an Overall
Measure of Risk (OMOR).
·
An effective framework for transitioning and refining concept
development in a multidisciplinary design optimization (MDO).
·
A means of using the results of first-principle analysis codes at
earlier stages of design.
This project develops a
process, tools and models implemented in Model Center, a general purpose design
environment and optimization program, to address these critical elements. Steps
in this process are:
1. Concept Exploration. A
multi-objective genetic optimization (MOGO) considers discrete major system
decisions and top level requirements including payload, choice of propulsion
system and power, range and speed, parent hull forms, hull materials, arrangement,
and manning. Genetic algorithms (GAs)
are able to explore a design space that is very non-linear, discontinuous, and
bounded by a variety of constraints and thresholds. These attributes prevent application of
mature gradient-based optimization techniques including Lagrange multipliers,
steepest ascent methods, linear programming, non-linear programming and dynamic
programming. GAs are also ideally-suited
for multi-objective optimization since they develop a population of designs
vice a single optimum. This population
can be forced to spread-out over a non-dominated frontier of design
alternatives as illustrated in Figure 1. A non-dominated solution, for a given
problem and constraints, is a feasible solution for which no other feasible
solution exists that is better in one attribute and at least as good in all
others. The non-dominated frontier is
the first product of this process.
Figure
1 - Two Objective Attribute Space
2. Customer selection of preferred design(s). There is no reason to pay more for the same
effectiveness or accept less effectiveness for the same cost. Preferred designs must always be on the
non-dominated frontier. The selection of
a particular non-dominated design depends on the decision-maker’s preference
for cost, effectiveness and risk. This
preference may be affected by the shape of the frontier and cannot be
rationally determined a priori. When
considering three attributes, the non-dominated frontier is a surface. Points on this surface represent feasible
ships, and can be mapped to specific design parameters. With such a surface, the full range of
cost-risk-effectiveness possibilities can be presented to decision-makers,
trade-off decisions can be made, and specific concepts can be chosen for
further analysis. “Knees in the curve” can be seen graphically. "Knees" are significant changes in
the slope of the frontier. It is often
desirable to be at the top of a high effectiveness to cost slope. Up to this point a little more cost will buy
a lot more effectiveness. Beyond it, the
cost of more effectiveness is much higher.
3. Concept Development. Starting
from selected concepts from Step 2, this step develops the selected concept
designs in a multidisciplinary design optimization (MDO) using mission
effectiveness, cost, risk or some weighted combination of these as a
single-objective attribute. Appropriate
constraints on effectiveness, cost and risk are included. Discrete design and requirement decisions
made in Step 2 become bounds and constraints in Step 3. This allows the application of more
traditional and efficient optimization methods.
Deliverables [1] through [9]
detail the research accomplished in Phase 1 of this project, based on the 3-step
process outlined above.
Task 1.0 Literature, Information and Data Search – Literature and data searches
were completed for the following topics: 1) Overall Measure of Effectiveness (OMOE)
and AHP validation [4]; 2) Risk approaches to naval ship design [1]; and 3)
Methods for design with uncertainty [1,2,3,5]. Reference lists and summaries
are included in these related theses.
Task 2.0
Multi-Objective Models and Probabilistic Design Optimization
Task 2.1 OMOE Model - A simplified methodology
was developed for building an Overall Measure of Effectiveness (OMOE) model
using the Analytical Hierarchy Process and Multi-Objective Value Theory [4,6].
A validation experiment was completed using the war-gaming software HARPOON3
[4], Figure 2. In this validation, ten student experts were
“educated” in a series of war-gaming experiences. OMOE functions were developed
using the OMOE methodology and expert opinion. The OMOE functions were then
used to rank a series of surface combatant designs. This ranking is compared to
the ranking results from a direct war-gaming comparison of the designs.
Figure 2 - Harpoon 3
Workspace
Figure 3 – Simulation Rank vs. Questionnaire Average Rank
An OMOE calculation using questionnaire averages most
closely matched and provided a good prediction of direct simulation results, Figure 3.
Task 2.2 Cost Model - Engineering cost models must be reliable, practical
and sensitive to the cost and performance impact of producibility enhancements. A
baseline surface combatant cost model was developed using a modified weight-based approach [4]. A more flexible model
will be developed in Phase 2 using ACEIT (Automated Cost Estimating Integrated
Tools). ACEIT is an automated architecture and framework for cost estimating.
It is a government-developed tool that has been used to standardize and
simplify the Life Cycle Cost estimating process in the government environment.
Core features include a database to store technical and (normalized) cost data,
a statistical package specifically tailored to facilitate cost estimating
relationship (CER) development and a spreadsheet that promotes structured,
systematic model development, and built-in government-approved inflation,
learning, time phasing, documentation, sensitivity/what-if, risk and other
analysis capabilities. Our task will be to adapt this general framework for
concept development naval ship cost analysis including producibility. Cost
uncertainty aspects will be integrated with Task 2.3.
Task 2.3 Risk and Uncertainty - The
DoD Risk Management Guide requires risk assessment of acquisition performance,
cost and schedule through the identification, subsequent analysis and
prioritization of adverse program events based on their probability of
occurrence and consequences. This type of risk assessment is very important in
concept exploration and design when considering new technologies, unique
processes and novel concepts. Uncertainty associated with the design process
itself and the definition and selection of specific design alternatives can
also have a significant impact on performance, cost and schedule risk.
Inherent, statistical and modeling uncertainty, and uncertainty due to human
error, must be considered in the design process, but uncertainty analysis
requires a more detailed and computationally intensive probabilistic approach.
It is most appropriate for post-exploration design optimization, after specific
cost and performance goals and thresholds have been set, to maximize the
probability of achieving these goals.
We have adopted a two-stage concept
design strategy that uses a multi-objective optimization and simplified risk event
approach for concept exploration, and a more rigorous multi-disciplinary
optimization with uncertainty for concept development. Concept exploration
identifies non-dominated design concepts and establishes the optimum
relationship between effectiveness, cost and risk given a broad selection of
technologies and design alternatives. Risk is defined using a separate
objective attribute, an Overall Measure of Risk (OMOR), which specifically
addresses the high-risk events associated with the selection of new technologies,
processes and concepts. With this perspective, decision-makers may establish
rational requirements, select technologies, narrow the design space, and
establish a non-dominated concept baseline design or set of designs. Once these
early decisions are made, concept development and the remaining design phases
add detail, refine requirements and reduce risk. Optimization continues into
concept development using uncertainty analysis with Confidence of Success (CoS)
as the third objective attribute. Tim Mierzwicki (MS Ocean Engineering 2003)
performed the initial risk and uncertainty literature search, developed the
OMOR approach, and performed an initial OMOR case study [1,7]
Since the mid 1990’s, Mavris
and associates at Georgia Tech have been exploring robust design techniques in
the presence of parameter uncertainties accounted for by assigning probability
distributions to selected model inputs.
The output responses have been generated by Monte Carlo simulations
performed either using the full model or a response surface approximation of
its output. They have also had success
using the Advanced Mean Value (AMV) method for calculating the cumulative
distribution function (CDF) of the model response in aerospace
multidisciplinary design applications.
The AMV method has the virtue of requiring far fewer model runs than
does a Monte Carlo simulation. It is one
of several Fast Probability Integration methods developed by Southwest Research
Institute and NASA Lewis Research Center.
The primary difference in our initial
research is that we are concentrating on the uncertainty generated from the
analysis process, the modeling uncertainty. For now, we assume the input
variables to be deterministic with randomness coming only from the embedded
uncertainties in the analyses.
Because of the
multidisciplinary nature of a ship synthesis model, perturbations introduced in
one analysis are carried forward to perturb the next and then on to the next
and so on. To further complicate the
issue, ship synthesis models typically require multiple iterations to balance
the design. Each subsequent level of
analysis may introduce its own uncertainty and inherit uncertainty from
previous analyses. It is difficult to
characterize this cascading of uncertainty through a highly nonlinear analysis.
We have examined several
methods to obtain information on the output distributions more efficiently than
through the Monte Carlo simulation. The
family of variance reduction techniques described, for example, by Law and
Kelton, are designed to obtain this information in a statistically efficient
manner. Statistical efficiency deals
with the precision of an estimator. If we are trying to describe an output
random variable (say the cost of a design), we are interested in certain
parameters that define its distribution; for example its mean and standard
deviation.
We used the Advanced Mean Value method to determine
statistics of the, now random, ship design characteristics that are calculated
by our model. We then determined
probabilities that a given design will have greater than any given level of
OMOE or the probability that the cost will be less than some level. We can also determine the probability that
the design will be feasible, i.e., it meets all the applicable
constraints.
The confidence of success is the joint probability
that a given design 1) is feasible, 2) that it will have a cost that is less
than a given maximum cost and 3) that it will have an OMOE that is greater than
a given required value. This confidence
of success (CoS) can be treated as a third objective function in a genetic
algorithm based optimization. Figure 4 is a three-dimensional Pareto frontier from which a
ship designer can pick the design of his choice considering both the overall
value of the design and the risk being taken at that point in the design
space.
This work was performed
primarily by Sandipan Ganguly (MS Ocean Engineering, 2002) and Emanuel Klasen
[5].
Figure 4 – Non-Dominated
(Pareto) Frontier with Confidence of Success
Task 2.4
Design Test Cases and Applications - The
approach, methods and tools developed in Phase 1 were exercised in a number of
case studies using a simple ship synthesis model [4,5,6,8,9], and the US Navy’s
Advanced Ship Synthesis and Evaluation Tool (ASSET) [2,3] in the ModelCenter
(MC) design environment. The simplified model case studies were performed
primarily by undergraduate ocean engineering students. A Mixed-Language
Programming (MLP) approach was used to interface with the ASSET software.
Component-based software construction has gained significant momentum
and become a main focus of software engineering research and computing. Even
though there are many standards available now for developing component-based
applications, there are still applications where a single-language based
approach is not suitable. Some of the actions that a program performs are best
expressed in a particular language, and the choice of a programming language is
strongly dictated by the programmer’s preference. The Mixed-Language
Programming (MLP) approach was used to build component-based software systems,
with a specific emphasis on the ship design problem [3]. This approach was
compared with a newer tool-based integration methodology of modeling and
building component-based software applications, using tools such as Phoenix
Integration’s ModelCenter and Analysis Server.
ModelCenter and ASSET were also applied to two ship design case
studies, LHA(R), a replacement for the US Navy amphibious assault ship, and
DDG-51, a destroyer class vessel [2,5]. Overall Measure of Effectiveness (OMOE)
and lead ship acquisition cost were the objective attributes. Design
feasibility was evaluated, and various ship parameter calculations were
performed using ASSET. ASSET was integrated with the design optimization
software DARWIN to obtain the non-dominated frontier over a range of acquisition
cost, Figure
5. Model Center software was used to integrate ASSET
and Darwin, Figure
6. VBScript components were used to run various ASSET
modules, apply the trade study option configurations, and calculate the
objective functions. Windows script components were developed to access the
operating system and invoke ASSET.
Figure
5 – LHAR
Non–Dominated Frontier
Figure
6 – DDG-51 ASSET
Model in Model Center
Phase
1 of this project accomplished the following objectives:
Phase 2 will include more Response
Surface Modeling (RSM), a more detailed Design of Experiments (DOE) and
variable screening, and capabilities for more physics-based modeling and
assessing the impact of new technologies. We would like to focus on multi-hull
high speed ships, and on using the computationally intensive tools required for
these ships including a Rankine panel method code for seakeeping and wave
resistance, an extensive dynamic simulation for machinery system definition and
performance, a structural finite element code and codes for accessing
structural vulnerability and survivability. We will also continue uncertainty
modeling and solution development in increasingly complex and realistic
problems with more demanding requirements for computational efficiency. Phase 2
will also include work on a surface ship manning model and submarine
applications.
[1] Mierzwicki, T. (2003), “Risk Index for
Multi-Objective Design Optimization of Naval Ships”, MS Thesis, Department
of Aerospace and Ocean Engineering, Virginia Tech, April 24, 2003.
[2] Neti, S.N. (2005), “Ship Design Optimization Using
ASSET”, MS Thesis, Department of Aerospace and Ocean Engineering, Virginia
Tech, February 10, 2005.
[3] Gunasekaran, Murali Krishnan (2003),
“Component-Based
Application Development Using a Mixed-Language Programming (MLP) Approach”,
MS Thesis, Department of Computer Science, Virginia Tech, December 2003.
[4] Demko, D. (2005), “Tools for Multi-Objective and
Multi-Disciplinary Optimization in Naval Ship Design”, MS Thesis,
Department of Aerospace and Ocean Engineering, Virginia Tech, May 2005.
[5] Klasen, E. (2006), “Confidence of
Success in Multi-Criteria Optimization of Multi-Disciplinary Ship Design Models”,
Report, Department of Aerospace and Ocean Engineering, Virginia Tech,
March 2006.
[6] Brown, A.J., Salcedo, J. (2003), "Multiple Objective Genetic Optimization In Naval Ship Design",
Naval Engineers Journal, Vol. 115, No. 4, pp. 49-61.
[7] Mierzwicki, T., Brown, A.J. (2004), “Risk Metric for Multi-Objective Design of Naval Ships”, Naval Engineers Journal, Vol. 116, No.
2, pp. 55-71.
[8] Good, N., Brown, A.J. (2006), “Multi-Objective Concept Design of an Advanced
Logistics Delivery Ship”, to be presented at ASNE Joint Sea Basing
Symposium, March 2006,
[9] Stepanchick, J., Brown, A.J. (2006),
“Revisiting
DDGX/DDG-51 Concept Exploration”, to be presented at ASNE Day, June 19-20,
2006, Arlington, VA.